Job absenteeism and health problems are frequently caused by elevated exposure to work-related stress. The public sector is particularly affected by this development. Nevertheless, public sector organizations seem to have issues to reliably detect stress or to discuss about this topic in an objective and factual manner. Data visualizations have been found to be a powerful boundary object for sense-making and for unraveling issues that lie under the surface. Based on a pilot study at a medium-sized municipality in Switzerland, we thus developed, tested, and discussed various alternative visual representations for creating awareness about occupational stress. The results of this study showcase the hidden potential and perils of analyzing physiolytics data on aggregate level.

Stress is a current issue in the workplace, manifesting itself through both psychological and physiological reactions. Biosensors might improve stress monitoring in the workplace, when employees become wearable device users. Yet, it remains unclear how to identify stress patterns through biosensors without direct observation of the users' activities. In particular, non-physiological aspects of employee activities altering physiological reactions, such as motion activity, may also be associated with stress measures. This longitudinal experimental study examines remote stress identification by testing whether a non-physiological signal of physical activity may improve the classification of stress-related physiological data collected through biosensors. The participants are 18 employees from Public Administration sector wearing biometric devices for around two months in the workplace. This study investigates the stress-related data classification, using established physiological measures (Galvanic Skin Response and Heart Rate) combined with a new non-physiological measure, associated with the user's physical activity (Motion Activity). Stress-related patterns are explored through unsupervised learning approach with help of Gaussian Mixture Model and K-Means classification analysis, completed by the bootstrap confidence intervals for evaluating uncertainty of classification. The results demonstrate that complementing physiological signals with a non-physiological signal, such as a physical activity-related information, improves stress pattern recognition through detection of emotional overarousal, arousal, and relaxation. These findings are especially promising in the context of the use of wearable devices for stress management, when stress-monitoring is done remotely and user' activity is not directly observed during measurements. Further research and cross-validation procedures should be used for building stress-identification algorithms for remote stress monitoring that include physiological and nonphysiological signals. Better understanding of stress measures may enhance the quality of stress management data collection processes through Information Systems, involved in the use of wearable devices in the workplace, and strengthen the data governance.

Statistical interpretation of stress-related indicators collected through wearable biosensors often relies on benchmarking, especially in the context of stress management interventions. However, it remains unclear how to construct stress level benchmarks for group stress-related indicators using limited historical data. This study examines whether the method of numerical simulation of stress-related responses could contribute to constructing benchmark curves. Experimental data consists of physiological and non-physiological signals of 18 Swiss public servants collected through wearable biosensors. This study draws upon Stress Pattern Recognition algorithm and Markov Chain modeling for simulating emotional responses according to specified data-driven scenarios of high and low stress. Proposed method allows constructing benchmark curves for an Overarousal Index. Results demonstrate that numerical simulation based on small datasets can be used effectively for constructing stress level benchmarks. The findings contribute to methodological knowledge in statistical learning on Stress Pattern Recognition algorithms and Markov Chains modeling by expanding their application to a new field of emotional response simulation according to scenarios.

No abstract